[BUG] Augmented assignments double-evaluate RHS expression #1230

Bug Description

The codegen for augmented assignments (x += expr, x *= expr, etc.) evaluates the RHS expression twice when the target is a simple scalar variable.

In emit_AugAssign, the RHS is evaluated at line 3348 via rhs = adj.eval(node.value), but for simple ast.Name targets this result is unused — the code falls through to make_new_assign_statement() which rewrites the augmented assignment as x = x + expr via a new ast.BinOp node, re-evaluating node.value a second time.

Reproducer:

  @wp.func
  def side_effect_add(counter: wp.array(dtype=int), a: float, b: float) -> float:
      wp.atomic_add(counter, 0, 1)
      return a + b

  @wp.kernel
  def test_kernel(counter: wp.array(dtype=int), result: wp.array(dtype=float)):
      total = float(0.0)
      total += side_effect_add(counter, 1.0, 2.0)
      total += side_effect_add(counter, 3.0, 4.0)
      result[0] = total

  counter = wp.zeros(1, dtype=int, device="cuda:0")
  result = wp.zeros(1, dtype=float, device="cuda:0")
  wp.launch(test_kernel, dim=1, inputs=[counter, result], device="cuda:0")
  print(counter.numpy()[0])  # Expected: 2, Actual: 4

Impact:

• Redundant loads and arithmetic in the generated CUDA/CPU code for any kernel using +=, *=, etc. on scalar variables
• Approximately doubles the instruction count for the RHS of each affected augmented assignment
• Particularly noticeable in tight loops where augmented assignments accumulate values (e.g., reductions, dot products)

System Information

No response